通信基站虚拟电厂能源智能调度优化：基于LSTM预测与强化学习决策的协同框架Smart Energy Dispatch for Telecommunication Base Station Virtual Power Plants: An LSTM–Reinforcement Learning Framework

Abstract

With the continuous increase in the integration level of renewable energy, virtual power plants (VPPs) have become an important mechanism for supporting grid flexibility and coordinating distributed energy resources. Accurate power scheduling is critical to realizing efficient energy management and minimizing operating costs for virtual power plants. This paper proposes a deep-learning-based power scheduling method for virtual power plant services of communication base stations. By integrating long short-term memory (LSTM) networks and reinforcement learning technologies, an integrated “prediction-decision” optimization framework is constructed. The LSTM network is adopted to mine the time-series dependence characteristics of power demand and achieve high-accuracy power demand forecasting. Reinforcement learning is applied to dynamically optimize the energy scheduling strategy of base stations to realize the optimal balance between reducing operating costs and improving energy sustainability. Experimental results demonstrate that the proposed model outperforms conventional methods in terms of prediction accuracy and operational efficiency. It can effectively cut the operating costs of virtual power plants and improve energy utilization efficiency. This study provides a prediction-decision framework for energy management of virtual power plants built upon distributed communication base-station resources and is of great significance for constructing a more resilient and sustainable new-type power system.

随着可再生能源整合程度的持续提升，虚拟电厂（VPPs）已成为保障电网稳定运行、优化能源资源配置的重要聚合机制。精准的电力调度是实现虚拟电厂高效能源管理和最小化运营成本的关键。本文提出了一种面向通信基站虚拟电厂服务的深度学习电力调度方法，通过融合长短期记忆（LSTM）网络与强化学习技术，构建“预测-决策”一体化优化框架：利用LSTM网络挖掘电力需求的时序依赖特性，实现高精度电力需求预测；基于强化学习动态优化基站能量调度策略，在运营成本与可再生能源利用之间进行动态权衡。实验结果表明，所提出的模型在预测精度和运营效率方面均优于传统方法，能够有效降低虚拟电厂运营成本并提升能源利用率。本研究为分布式基站资源虚拟电厂的能源管理提供了一种预测—决策协同框架，对构建更具韧性和可持续性的新型电力系统具有重要意义。